Prudent walks and polygons - CiteSeerX

IOP PUBLISHING

JOURNAL OF PHYSICS A: MATHEMATICAL AND THEORETICAL

J. Phys. A: Math. Theor. 42 (2009) 095205 (16pp)

doi:10.1088/1751-8113/42/9/095205

Prudent walks and polygons Timothy M Garoni, Anthony J Guttmann, Iwan Jensen and John C Dethridge ARC Centre of Excellence for Mathematics and Statistics of Complex Systems, Department of Mathematics and Statistics, The University of Melbourne, Victoria 3010, Australia E-mail: [email protected], [email protected] and [email protected]

Received 17 October 2008, in final form 3 January 2009 Published 4 February 2009 Online at stacks.iop.org/JPhysA/42/095205 Abstract We have produced extended series for two-dimensional prudent polygons, based on a transfer matrix algorithm of complexity O(n5 ), for a series of n-step polygons. For prudent polygons in two dimensions we find the growth constant to be smaller than that for the corresponding walks, and by considering three distinct subclasses of prudent walks and polygons, we find that the growth constant for polygons varies with class, while for walks it does not. We give exact values for the critical exponents γ and α for walks and polygons, respectively. We have extended the definition of prudent walks to three dimensions and produced series expansions, using a back-tracking algorithm, for both walks and polygons. In the three-dimensional case we estimate the growth constant for both walks and polygons and also estimate the usual critical exponents γ , ν and α. PACS numbers: 05.50.+q, 05.70.Jk, 02.10.Ox

1. Introduction A well-known long standing problem in combinatorics and statistical mechanics is to find the generating function for self-avoiding polygons (or walks) on a two-dimensional lattice, enumerated by perimeter. These models are of considerable physical significance, describing as they do the behavior of long-chain polymers in dilute solution and biological vesicles. As a consequence, thousands of papers have been devoted to aspects of their behavior, and yet we still lack an exact solution for any relevant generating function. Let F (z) be a formal power series in z with coefficients in C. It is said to be differentiably finite or D-finite if there exists a non-trivial differential equation: Pd (z)

dd d F (z) + · · · + P1 (z) F (z) + P0 (z)F (z) = 0, d dz dz

1751-8113/09/095205+16$30.00 © 2009 IOP Publishing Ltd Printed in the UK

(1) 1

T M Garoni et al

J. Phys. A: Math. Theor. 42 (2009) 095205 700

600

500

400

300

200

100

0 0

100

200

300

400

500

600

Figure 1. Typical prudent walk of n = 2000 steps, generated via Monte Carlo simulation using a pivot algorithm.

with Pj a polynomial in z with complex coefficients [1, 2]. Recently, we have gained a greater understanding of the difficulty of this problem, as Rechnitzer [3] has proved that the (anisotropic) generating function for square lattice self-avoiding polygons is not differentiably finite (D-finite), confirming a result that had been previously conjectured on numerical grounds [4]. The definition of D-finite functions can be extended beyond functions of one variable. Let G(x, y) be a formal power series in y with coefficients that are rational functions of x. Such a series is said to be D-finite if there exists a non-trivial differential equation: Qd (x, y)

∂d ∂ G(x, y) + · · · + Q1 (x, y) G(x, y) + Q0 (x, y)G(x, y) = 0, d ∂y ∂y

(2)

with Qj a polynomial in x and y with complex coefficients [2]. There are many non-trivial simplifications of the self-avoiding walk or the polygon problem that are solvable [5], but all the simpler models impose an effective directedness or an equivalent constraint that reduces the problem, in essence, to a one-dimensional problem, and all such problems are known to have D-finite solutions. Prudent walks were introduced to the mathematics community by Pr´ea in an unpublished manuscript [6] and more recently have been reintroduced by Duchi [7]. A prudent walk is a connected path on the square lattice Z2 such that, at each step, the extension of that step along its current trajectory will never intersect any previously occupied vertex. Such walks are clearly self-avoiding. We enumerate prudent walks by the number of steps n. We take the empty walk of length zero, given by the vertex (0, 0), to be a prudent walk. The definition of prudent walks can be generalized to hyper-cubic lattices Zd . Figure 1 shows a typical prudent walk of n = 2000 steps, generated via Monte Carlo simulation using a pivot algorithm [8]. 2

J. Phys. A: Math. Theor. 42 (2009) 095205

T M Garoni et al

Note the roughly linear behavior—it is believed, although unproven, that the mean-square end-to-end distance grows like n2 for prudent walks, i.e. that ν = 1. The bounding box of a prudent walk is the minimal rectangle containing the walk. The bounding box may reduce to a line or even to a point in the case of the empty walk. One significant feature of two-dimensional prudent walks is that the end-point of a prudent walk is always on the boundary of the bounding box. Each step either lies along the boundary perimeter or extends the bounding box. Note that this is not a characterization of prudent walks. There are walks such that each step lies on the perimeter of the bounding box that are not prudent. We call such walks perimeter walks, and they will be the subject of a future publication [9]. The simplest example of a perimeter walk which is not a prudent walk is the , where the last west step breaks the prudent walk with steps NEESW, that is the walk, restriction since it steps in the direction of the occupied vertex at the origin. Furthermore, if one extends the definition of prudent walks to three-dimensional walks, it is not true that each step of the walk lies on the perimeter of the bounding box. Again, one can define three-dimensional walks with the property that each step lies on the perimeter of the bounding box, and these too will be discussed in the aforementioned publication [9]. Another feature of prudent walks that should be borne in mind is that they are, generally speaking, not reversible. If a path from the origin to the end-point defines a prudent walk, it is unlikely that the path from the end-point to the origin will also be a prudent walk. Ordinary SAW are of course reversible. A related, but not identical, model was proposed more than 20 years ago in the physics literature [10], where it was named the self-directed walk. In [10] the authors conducted a Monte Carlo study and found that ν = 1, where R 2 n ∼ cn2ν . Here R 2 n is the mean square end-to-end distance of a walk of length n. They also sketched an argument that the critical exponent γ , characterizing the divergence of the walk generating function C(x) =  cn x n ∼ A(1 − μx)−γ , should be exactly 1, corresponding to a simple pole singularity. This model differs from prudent walks in that different probabilities are assigned to different walks of equal length, depending on the number of allowable choices that can be made at each step. For the problem of prudent walks, all realizations of n-step walks are taken to be equally likely. The problem proposed by Pr´ea was subsequently revived by Duchi [7] who also studied two proper subsets, called prudent walks of the first type and prudent walks of the second type (see figure 2 for examples). Prudent walks of the first type are prudent walks in which it is forbidden for a west step to be followed by a south step, or a south step to be followed by a west step. Equivalently, prudent walks of the first type must end on the northern or eastern sides of the bounding box. Such walks are sometimes referred to as two-sided prudent walks. Prudent walks of the second type are prudent walks in which it is forbidden for a west step to be followed by a south step when the walk visits the top of its bounding box and a west step to be followed by a north step when the walk visits the bottom of its bounding box. Equivalently, prudent walks of the second type must end on the northern, eastern or southern sides of their bounding box. Such walks are sometimes referred to as three-sided prudent walks. Duchi found the solution for prudent walks of the first type and gave functional equations for the generating function of (unrestricted) prudent walks. More recently the problem has been revisited by Bousquet-M´elou [11], who gave a systematic treatment of all three types, and in particular gave a solution for the generating function for prudent walks of the second type. Numerical results for prudent walks (unrestricted) can be found in [12]. In the following section, we give a description of the transfer matrix algorithm used to enumerate prudent walks and polygons. In section 3 we give a summary of known results 3

T M Garoni et al

J. Phys. A: Math. Theor. 42 (2009) 095205

(a)

(b)

(c)

(d )

Figure 2. Typical examples of (a) a type-1 prudent walk, (b) a type-2 prudent walk, (c) a general prudent walk and (d) a prudent polygon.

for prudent walks including expressions for the exact generating functions for prudent walks of the first and second kinds as well as a functional equation for the unrestricted problem. Section 4 gives a summary of exact results for two- and three-sided prudent polygons and numerical results from the analysis of the unrestricted prudent polygon generating function. In section 5 we study prudent walks and polygons on the cubic lattice. We give a brief description of the back-tracking algorithm used to calculate the series for the generating functions and analyze these series to obtain numerical estimates for the critical points and exponents α, γ and ν. Finally in section 6 we summarize our findings. 2. Computer enumeration In two dimensions, we use transfer-matrix algorithms to count the number of prudent walks and polygons. Consider the enumeration of n-step unrestricted prudent walks. Any partially completed walk of say m < n steps is contained within a bounding box, which is the smallest rectangle into which we can fit the walk. It is easy to show that the end-point of the walk must be on a side of this box. The next step must either move away from the box, making it larger, or along the current side, which is possible only if the walk has not already visited vertices lying in that direction. The transfer-matrix algorithm proceeds by adding a single step to the prudent walk at a time. Given a prudent walk, only a small amount of information about the walk is required in order to determine how to extend the walk by a single step so as to form a longer prudent walk. This information is called a configuration. Any partially completed prudent walk corresponds to a unique configuration and there is only a finite number of such configurations. The information needed to describe the configuration of prudent walks is the dimensions of the bounding box, which side of the box the walk end-point is on, the location of the end-point on that side and an integer representing the directions in which the walk is allowed to be extended, i.e. whether there are sites of the walk along the side to the left of the end, to the right of the end or neither (both is not possible.) For a given configuration, there are three possible steps a walk can take: • a step outward away from the current side; • a step to the left along the current side, if there are no walk sites in that direction; • a step to the right along the current side, if there are no walk sites in that direction. When we move from a side to a corner of the bounding box, we consider the end point of the walk to have moved to the other side only if that side has been moved outward by the step. 4

T M Garoni et al

J. Phys. A: Math. Theor. 42 (2009) 095205

Starting from a walk of size 1 the algorithm progresses by adding a single step at a time until a given maximal length n is reached. During this computation we count the number of walks of intermediate length m < n corresponding to each possible configuration. Any given configuration thus corresponds to an equivalence class of partially completed prudent walks, and when adding a single step to each configuration we are add, a step to each walk in the equivalence class. We keep a list of the possible configurations and the number of walks in the equivalence class of that configuration. The algorithm steps through the list, and for each source configuration s, generates the new target configurations tj that can be obtained by adding a single step (there are at most three new target configurations tj per source configuration). We then add the number of walks cs with the source configuration s to the number of walks ctj of each of the target configurations tj . The length of a walk is equal to the number of iterations of the algorithm and the total number of walks of a given length is found by summing the number of walks cs in the list of source configurations. Without loss of generality, we can assume that the end-point of the walk is on the top side of the bounding box, since direction is not important for unrestricted prudent walks. We can also assume that the end of the walk is not farther from the left side than from the right side, since the number of ways in which a walk can be completed is invariant under reflection. For prudent polygons, the position of the end-point of the walk is not enough information to determine if the walk can be extended to create a prudent polygon. The position of the start-point of the walk is also required. So we have to extend a configuration to include the coordinates of the start-point of the walk. A prudent walk can only be extended to form a prudent polygon if the starting point of the walk either lies on one of the sides of the bounding box or is adjacent to a site on the current side of the bounding box that is in a direction in which the walk can extend. Configurations that do not satisfy this restriction can be discarded since such walks can never lead to prudent polygons. If we are enumerating polygons of size n, and for a given configuration of m-step walks we cannot reach a site adjacent to the end-point in n − m steps or less, we can ignore that configuration. When calculating the number of polygons of a given size, we sum the number of walks only for those configurations s where the end-point is adjacent to the start-point. The number of configurations grows like n4 . The width w of the bounding box can vary from 1 to n while the length l can vary from 1 to n–w. Some smaller boxes cannot occur because for w × l < n the walk cannot fit within the box, but we shall ignore this effect for simplicity. In a box of size w × l the end-point can be in any of the w positions on the top edge while the starting point can be in 3w + 2l positions, so up to constant factors (arising from the restrictions on the directions the walk may take etc) we have that the number of configurations must be proportional to n n−w   w=1 l=1

w(3w + 2l)
4.4107, while the second iterates increased more slowly and gave μ2P > 4.4128. The third iterates reached a maximum at 116 terms, and then a minimum at 201 terms and then steadily increased. If this monotone trend continues, we can conclude μ2P > 4.4136. Combining all these methods, we estimate μ2P = 4.415 ± 0.001 or μP = 2.1012 ± 0.00025 which encompasses all the results. In figure 3 we show a plot of the ratios of successive terms plotted against 1/n. Assuming

2−α the singular part of the generating function behaves as Ps (x) ∼ const. 1 − μ2P x , then asymptotically, the ratios should approach μ2P , with gradient μ2P (α − 3). From the ratio plot, a limit around 4.415 is very plausible, though there is still a small amount of curvature in the locus of ratios. This is almost certainly due to the presence of exponentially small corrections due to one or more nearby singularities on the real axis. Extrapolating this locus to the estimated limit μ2P ≈ 4.415, we estimate the slope in the vicinity of the y-axis to be −20, so that α ≈ −1.5. We have also used biased differential approximants with the critical point biased at μ2P = 4.415, and while this signaled a confluent singularity at the critical point, it gave a consistent value for the exponent, α − 2 as −3.5 ± 0.1, so that again α ≈ −1.5. In summary, we find for the singular part of the generating function, Ps (x) ∼

2−α const. 1 − μ2P x , where now α ≈ −1.5. We do not quote error bars as this estimate 11

T M Garoni et al

J. Phys. A: Math. Theor. 42 (2009) 095205 4.42 4.40

pn/pn-1

4.38 4.36 4.34 4.32 4.30

0

0.002

0.004

0.006

0.008

0.010

1/n Figure 3. Plot of successive ratios of terms pn /pn−1 against 1/n for prudent polygons of halfperimeter n. Our best estimate for the critical point μ2P = 4.415 is marked with a large dot on the y-axis.

of α is very sensitive to the estimate of the critical exponent used to bias the results. If we bias the critical point at μ2P = 4.416, a change of only 1 in the least significant digit, the exponent estimate from differential approximants changes to −4.2 ± 0.02. Thus our estimate of α can only be taken as indicative rather than precise. What is clear however is that the singularity is unlikely to be of square-root type, as is the case for types 1 and 2 polygons, but of course there is no obvious reason why it should be. For the anisotropic case, we write the generating function as   pm,n x m y n = Hn (x)y n , P (x, y) = m,n

n

where the first few Hn (x) are 2x H1 (x) = (1 − x) x(2 + 2x − x 2 ) H2 (x) = (1 − x)3 x(2 + 9x + 6x 2 + x 4 + x 5 ) H3 (x) = (1 − x)5 (1 + x) x(2 + 19x + 42x 2 + 37x 3 + 18x 4 + 21x 5 + 11x 6 − x 8 ) H4 (x) = (1 − x)7 (1 + x)2 V14 (x) , H5 (x) = (1 − x)9 (1 + x)4 (1 + x + x 2 ) V22 (x) H6 (x) = , 11 (1 − x) (1 + x)6 (1 + x + x 2 )2 V32 (x) H7 (x) = , (1 − x)13 (1 + x)8 (1 + x + x 2 )4 (1 + x 2 ) where Vi (x) denotes a polynomial of degree i. From the above, we see the relentless buildup of cyclotomic polynomials of increasingly high order. As explained above, if this pattern persists, the underlying generating function cannot be D-finite. While this does not a priori prove that 12

J. Phys. A: Math. Theor. 42 (2009) 095205

T M Garoni et al

the isotropic generating function is not D-finite, we know of no combinatorial problem where this is the case. That is to say, where the isotropic generating function is D-finite, while the anisotropic generating function is not. 5. Three-dimensional prudent walks and polygons We have enumerated all prudent walks of up to n = 23 steps on the three-dimensional simple cubic lattice using a simple backtracking algorithm. As for the case of ordinary SAW, it is a far more difficult task to perform enumerations in three dimensions than it is in two. To take advantage of the inherent symmetry in the problem we only explicitly counted walks whose first step was in the positive x direction, whose first step out of this line (if any) was in the positive y direction and whose first step out of this plane (if any) was in the positive z direction. In addition, we trivially parallelized the backtracking algorithm: In three dimensions there are 16 four-step prudent walks whose first two steps are precisely East then North. We independently enumerated the completions to n-step walks of each of these prefixes. Similarly there are 4 four-step prudent walks whose first three steps are East, East, North. Again we independently enumerated the n-step completions of each of these prefixes. Finally, the prudent walks whose first three steps are East, East, East were all enumerated together in the same computation. Thus, for each value of n we ran a total of 21 independent computations. The enumerations for n = 23 took a total of around 2377 h (roughly 100 h per independent computation). The computations were performed on tango, a 95-node Linux cluster at the Victorian Partnership for Advanced Computing (VPAC). Each node consists of two AMD Barcelona 2.3 GHz quad core processors. In principle, it is probably possible to obtain another term or two by applying the two-step method [18], but we have not pursued this here. For each 1  n  23 we computed the number of prudent walks, cn , the number of prudent polygonal returns, un+1 and the sum of squared end-to-end distances cn R 2 n , summed over all prudent walks. The results of our enumerations are presented in table 1. We analyzed the various series by the method of differential approximants, as well as variants of the ratio method, as used in the analysis of the polygon generating function in the previous section [17]. For the prudent walk generating function, for which we expect  C(x) = cn x n ∼ const. (1 − x/xc )−γ , we found a singularity on the positive real axis at xc = 0.22265 ± 0.0003 with the corresponding exponent γ = 1.68 ± 0.03. Biasing the value of the critical point at the central estimate, that is setting xc = 0.22265 gives the corresponding biased estimate of γ = 1.67 from first-order differential approximants and γ = 1.68 from second-order differential approximants. For SAW, the analogous values are xc (SAW) ≈ 0.2134907 and γSAW ≈ 1.1567, so prudent walks are exponentially rare among SAW, and the two models have different critical exponents. Additionally, for three-dimensional prudent walks we find that there is a singularity in the generating function C(x) on the negative real axis that appears to be at, or just beyond, x = −xc . For the corresponding self-avoiding walk model, it is known that there is a singularity (the analog of an anti-ferromagnetic singularity for a magnetic model) exactly at x = −xc (SAW), but the argument for the location of that singularity in the case of SAW does not translate to prudent walks. The corresponding exponent is, very approximately, the same magnitude as that of the physical singularity, but opposite in sign. That is to say, there appears to be a singularity of the form ∗

const. (1 + x/x ∗ )γ , 13

T M Garoni et al

J. Phys. A: Math. Theor. 42 (2009) 095205

Table 1. Exact enumeration data on the simple cubic lattice. Here cn denotes the number of n-step prudent walks, un+1, the number of (n + 1)-step prudent polygonal returns and R 2 n the average squared end-to-end distance of n-step prudent walks.

n

cn

cn R 2 n

un+1

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23

1 6 30 150 726 3510 167 34 795 18 375 246 176 6382 827 8638 387 213 66 180 556 206 840 524 742 390 386 6526 181 067 988 30 838 327 781 10 387 690 560 718 179 033 006 5854 825 952 831 5558 380 594 975 182 14 175 228 328 442 174 805 959 153 119 262 370 427 057 572 4550

0 6 72 582 4032 255 42 153 048 881 118 492 5616 269 099 34 144 356 280 762 839 334 398 106 4368 205 560 008 22 105 173 637 672 533 839 505 646 269 076 118 6608 134 784 799 054 86 671 428 938 557 52 332 807 521 103 670 164 221 451 827 7040 807 024 661 037 2494 395 111 666 883 222 48 192 780 251 992 208 934

0 0 0 24 0 240 0 2544 0 318 00 0 435 864 0 632 3352 0 956 471 04 0 149 393 4516 0 239 340 016 00 0 391 427 518 152 0 651 194 900 1648

where x ∗  xc and γ ∗ ≈ γ , as well as the physical singularity. For two-dimensional prudent walks, there is also evidence of a singularity on the negative real axis, but located considerably further away from the origin than the physical singularity. That is to say, x ∗ (2d) > xc (2d). To calculate the exponent ν characterizing the mean-square end-to-end distance, R 2 n ∼ const. n2ν we analyzed the series for the sum of the squared end-to-end distances, which diverges at xc with exponent γ + 2ν. From a differential approximant analysis we found xc ≈ 0.22265 with γ + 2ν ≈ 3.20, so that ν ≈ 0.76. We also analyzed the series R 2 n directly and obtained an estimate for ν consistent with that just quoted, but less precise. The polygonal returns un include a factor 2n corresponding to n possible starting points, and a factor of 2 as the path may be traversed clockwise or anticlockwise. We have analyzed the series with coefficients pn = un /2n. We only have 11 coefficients, which is not really enough for any but the crudest analysis.  From a differential approximant analysis, we find that the generating function P (z) = n pn zn is singular at z = zc = 0.0499 ± 0.0006 with an exponent 2.3 ± 0.5. That is to say, Ps (z) ∼ const. (1 − z/zc )2.3 . Note that from our analysis of the corresponding walk series, xc2 = 0.04957 ± 0.00014. Thus from the numerical data, it is entirely possible that zc = xc2 for three-dimensional prudent polygons, just like the analogous situation for three-dimensional SAW, but unlike the situation for two-dimensional prudent 14

J. Phys. A: Math. Theor. 42 (2009) 095205

T M Garoni et al

walks and polygons, for which, as discussed above, prudent polygons are exponentially rare among prudent walks. In two dimensions, the condition of prudency severely limits the number of walks that are also polygons, as the prudency constraint tends to drive the end-point away from the origin. In three dimensions, the prudency constraint is much less severe. In particular, the end-point of the walk does not have to be on the surface of the minimum bounding box. However we have been unable to prove that three-dimensional prudent walks and polygons have the same growth constant. In particular, the proof of the corresponding result for standard SAW and SAP does not carry over to the prudent case. In higher dimensions it is increasingly likely that prudent walks and polygons have coincident growth constants, though we have no numerical data to investigate this point. If it is true that zc = xc2 for three-dimensional prudent polygons, then we can carry out a biased analysis in which we fix zc at xc2 . In that case we find that the estimate for the exponent 2 − α of the polygon generating function is a little higher at 2.5 ± 0.2 so that α = −0.5 ± 0.2 6. Conclusion We have defined and analyzed series for prudent walks and polygons in both two and three dimensions. In two dimensions, we also discussed two subsets, which are exactly solvable. We found that two-dimensional prudent polygons are exponentially rare among prudent walks, unlike the situation for two-dimensional SAW. We gave numerical arguments in support of the conjecture that the anisotropic generating function for two-dimensional prudent polygons is not D-finite. We derived extensive series for three-dimensional prudent walks and polygons. As far as we are aware, this problem has not been studied previously. We have given estimates of the critical point and critical exponents. In terms of the usual notation we found that the growth constant for walks is μ ≈ 4.491, with exponents γ ≈ 1.68, ν ≈ 0.76 and α ≈ −0.3 based on an unbiased estimate of the critical point. For SAW μSAW ≈ 4.68404, γ ≈ 1.1567, ν ≈ 0.5875 and α ≈ 0.2375. For SAW we have the hyper-scaling relation dν = 2 − α. While there is no a priori reason to expect this hyper-scaling relation to hold for three-dimensional prudent walks, as it is not a statistical mechanical model derived from a Hamiltonian, we nevertheless note that 3ν ≈ 2.28 while 2 − α ≈ 2.28. The uncertainties in our exponent estimates are too great to attach much significance to this approximate equality, except to flag it as a possibility. In terms of physical significance, the model of prudent walks, while exponentially rare among SAW in both two and three dimensions, is nevertheless exponentially abundant compared to any other solved or numerically estimated model. Other variants of the model that relax the prudency constraint to some extent are likely to have growth constants even closer to that of SAW, and these will be the subject of a future publication [9]. Acknowledgments We would like to thank Simone Rinaldi and Enrica Duchi for introducing us to this problem, and Mireille Bousquet-M´elou for several enlightening discussions and access to her results prior to publication. Similarly, we thank Uwe Schwerdtfeger for provision of his results prior to publication. We are also grateful to Andrew Conway, who checked much of the data independently, and to the referees whose queries resulted in a significantly improved manuscript. The calculations presented in this paper were performed on the facilities of the Australian Partnership for Advanced Computing (APAC) and the Victorian Partnership 15

J. Phys. A: Math. Theor. 42 (2009) 095205

T M Garoni et al

for Advanced Computing (VPAC). We gratefully acknowledge financial support from the Australian Research Council. References [1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11] [12] [13] [14] [15] [16] [17] [18]

16

Stanley R P 1980 Differentiably finite power series Eur. J. Comb. 1 175–88 Lipschitz L 1989 D-finite power series J. Algebra 122 353–73 Rechnitzer A 2003 Haruspicy and anisotropic generating functions Adv. Appl. Math. 30 228–57 Guttmann A J and Conway A R 2001 Square lattice self-avoiding walks and polygons Ann. Comb. 5 319–45 Bousquet-M´elou M 1996 A method for the enumeration of various classes of column-convex polygons Disc. Math. 154 1–25 Pr´ea P 1997 Exterior self-avoiding walks on the square lattice (unpublished) Duchi E 2005 On some classes of prudent walks In Proc. FPSAC’05 Taormina, Italy Madras N and Sokal A D 1988 The pivot algorithm: a highly efficient Monte Carlo method for the self-avoiding walk J. Stat. Phys. 50 109–86 Garoni T M and Guttmann A J 2009 Perimeter and quasi-prudent self-avoiding walks and polygons (in preparation) Turban L and Debierre J-M 1987 Self-directed walk: a Monte Carlo study in two dimensions J. Phys. A: Math. Gen. 20 679–86 Bousquet-M´elou M 2008 Families of prudent self-avoiding walks, arXiv hal-00276681 Dethridge J C and Guttmann A J 2008 Prudent self-avoiding walks Entropy 10 309–18 Flajolet P and Sedgewick R 2008 Analytic Combinatorics (Cambridge: Cambridge University Press) Stanley R P 1999 Enumerative Combinatorics vol 2 (Cambridge: Cambridge University Press) Bousquet-M´elou M and Rechnitzer A 2002 Lattice animals and heaps of dimers Discrete Math. 258 235–74 Schwertdfeger U 2008 Exact solution of two classes of prudent polygons, arXiv math.CO 0809.5232 Guttmann A J 1989 Asymptotic analysis of coefficients Phase Transitions and Critical Phenomena vol 13 ed C Domb and J L Lebowitz (London: Academic) 1–234 Clisby N, Liang R and Slade G 2007 Self-avoiding walk enumeration via the lace expansion J. Phys. A: Math. Theor. 40 10973–1017